Apply optimizations and fixes to Enumerable.Take(Range)

[eiriktsarpalis]

Makes the following changes to the new Enumerable.Take(Range) implementation:

• Separate SizeOpt and SpeedOpt implementations for non-fromEnd ranges.
• Optimize fromEnd ranges using Enumerable.TryGetNonEnumeratedCount to obtain the source count.
• Change enumerator disposal semantics to more closely follow the TakeLast and SkipLast implementations.
• Finally, expresses the TakeLast and SkipLast in terms of the Take(Range) implementation.

Fix#48631.

[ghost]

Tagging subscribers to this area: @eiriktsarpalis
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Makes the following changes to the new Enumerable.Take(Range) implementation:

• Separate SizeOpt and SpeedOpt implementations for non-fromEnd ranges.
• Optimize fromEnd ranges using Enumerable.TryGetNonEnumeratedCount to obtain the source count.
• Change enumerator disposal semantics to more closely follow the TakeLast and SkipLast implementations.
• Finally, expresses the TakeLast and SkipLast in terms of the Take(Range) implementation.

Author:	eiriktsarpalis
Assignees:	eiriktsarpalis
Labels:	area-System.Linq
Milestone:	6.0.0

[eiriktsarpalis]

cc @Dixin

[eiriktsarpalis]

I've been investigating the seemingly unrelated test failure affecting mono debug builds. The issue is caused by xunit's generic test parameter resolution logic: the removal of the private SkipLastIterator method seems to result in incorrect generic parameters being resolved for certain generic theory parameters.

I have pushed a commit to demonstrate this and to further show that the failures are not due to a regression introduced by this PR, but rather a bug of xunit running over mono debug builds.

[eiriktsarpalis]

Hi @stephentoub@layomia, this is ready for review, would you be able to take a look?